Ten biases, one crowd. And the model that lies about both.
Each pair below is the same population measured twice — steered two opposite ways by nothing but context. You can't predict the single person, but you can move the tribe. Then switch on the model's fitted bell and watch the deeper trap: a tool that assumes a bell curve will draw one over anything — so its clean μ and σ can never tell you that the data was steered, or that the shape was never normal to begin with.
0%
◀ shows the trait lessaverageshows it more ▶
measured crowd unsteered baseline the model's fitted "normal"
The two context levers
Framing, anchors, mood, a headline. A transient tilt that bites the first pegs and fades.
The architecture's standing tilt — an opt-out, a layout, a path of least resistance. Constant at every peg, it never fades.
How independent each choice is. Couple them positively (chase streaks) or negatively (revert to the mean) and the bell stops being a bell.
Peg shape & ball size
How hard each choice tips. Sharper pegs throw further and hollow out the centre.
How alike the individuals are. Bigger, more varied balls widen and roughen the crowd.